Why YouTube's "Not Interested" doesn't work (and what does)

You've pressed it a hundred times. Nothing changed.

Most people who've used YouTube for more than a year have a similar story: they noticed the recommendations were getting worse, found the Not Interested and Don't recommend channel buttons, used them aggressively for a couple of weeks, and watched their recommendations slowly drift back to the same kind of content. The buttons worked at first. Then they didn't.

That's not a bug. It's how the system is designed to behave, given the metric YouTube optimises for.

What the button actually does

When you press Not Interested, three things happen:

1. Per-video suppression: that specific video stops appearing in your home feed. (Won't reappear unless it virally explodes and overrides personalisation.) 2. Per-channel signal (only on the "Don't recommend channel" sub-option): that channel's videos are de-weighted in your recommendations. 3. Implicit category signal: a small, decaying nudge against videos that share metadata features with the one you flagged.

What it does not do:

• It doesn't reset your overall recommendation model
• It doesn't generalise from "I hate clickbait" to "stop showing me clickbait"
• It doesn't override the algorithm's strongest signals (watch time on similar videos from your history)

That last point is the critical one. The algorithm's primary input is what you actually watched. Not Interested is a secondary input — much weaker than the action of clicking and watching even 15 seconds of a similar video three weeks ago.

Why it decays

YouTube's recommendation system optimises for expected watch time per session. Every decision it makes — what to put on your home feed, what to autoplay next, what thumbnails to show — is graded on whether you watched. Not Interested signals reduce one specific candidate pool, but the system has roughly five million other candidates. It will surface something else from the same engagement-rich bucket.

If you mark 100 clickbait videos as "Not Interested," the algorithm learns "this user dislikes these 100 specific channels." It does not learn "this user dislikes clickbait." The pattern at the abstraction level above the channel — the content class — is not what the model is structured around.

This is why the relief is short-lived. You suppress the supply you've already seen; the demand-side optimisation finds new supply.

What does work

Three interventions that actually shift the recommendations long-term, ordered by leverage:

1. Pause and clear your watch history

Settings → Your data in YouTube → Manage activity → Delete activity by → All time. Then turn History pause on.

This is the nuclear option, but it works because it removes the primary signal the algorithm uses. Within 24 hours your home feed will reset to "popular in your country" — generic, often boring, but no longer optimised against you. From there, only watch things you'd be glad you watched. Skip the things you'd be embarrassed by.

The cost: no resume-where-you-left-off, no recommendation graph for your real interests, no Watch Later context. The benefit: a clean baseline you can re-train from scratch.

2. Use the Subscriptions tab as your home

YouTube's Subscriptions tab shows only what your subscribed channels uploaded, in chronological order. It is not algorithmically optimised in the same way the home feed is — it's a much closer approximation of "what the people I chose to follow are doing right now."

If you use the Subscriptions tab as your default surface for a week, two things happen: you watch less YouTube (because the supply runs out faster), and you watch a higher share of stuff you'd rate positively. Both are the goal.

This is the single highest-effort-to-impact ratio for most people.

3. Filter the recommendations at the browser level

If you can't bring yourself to abandon the home feed (some people do use it productively for discovery), the tractable fix is to run a content-classifier filter over it. Browser extensions like PureFeed read each video card's title and description, classify the content (educational, entertainment, news, etc.), score it on dimensions like sensationalism and usefulness, and hide the ones above your sensationalism threshold.

The mechanism difference vs. Not Interested:

• Not Interested sends a feedback signal to YouTube's recommendation model, which weakly weighs it against many stronger signals
• A page-level filter reads the actual text of each recommendation as it's shown, and hides it before you scroll past it

The filter doesn't need the algorithm to cooperate. It works regardless of what the recommendations are because it acts at the rendering layer, not the model layer. Catches new clickbait the moment it appears; catches channels you've never seen; persists across browsing sessions without decay.

What else doesn't work

A few interventions people commonly try that fall in the same "feels like it should work" trap:

• Closing the YouTube tab faster: doesn't reduce watch time enough to outweigh the click signal that was already sent when you opened the video.
• Using Incognito: works only for that session. Sign back in and you're back where you started.
• "Following more good channels": improves your Subscriptions tab, doesn't fix the home feed (because the home feed weights subscriptions only a fraction of what it weights recommendations).
• "Marking as offensive": this is a moderation flag, not a personalisation signal. Use sparingly and only for content that actually violates terms; it's not a personal preference channel.

The honest summary

The button feels like it should work because the user interface implies it does. It says "Not Interested," you assume that means the algorithm now knows you're not interested in that kind of thing. What the algorithm actually heard is "not this specific video, but I'm still here, what else have you got."

If you want recommendations to actually change, you have two tractable paths: starve the algorithm of your watch data (option 1), or stop letting the algorithm pick your feed in the first place (options 2 and 3). The button is a third path that costs almost nothing to use, so use it — but use it knowing that the signal it sends is small compared to the signal you sent yesterday by watching three more clickbait videos.